ci: cache repo mirror, not workspace, expand docs

Working with the new documentation CI pipeline has uncovered problems caused by attempts to cache the entire repo workspace. Initially there are no problems but as pipelines are testing subsequent iterations of a branch, history may become non-linear. In one case repo refused to synchronize (with repo sync --force) the updated repository citing lack of common history.

In the experimental Oniro pipeline I've since switched to keeping a cache of the git repo miror, not repo workspace. This also helps with cache hit when multiple releases are being tested, as cache is mostly static (per release) and subsequent pielines do not fight with each other by ovrewriting the cache to their preferred state.

Some variable names are updated to better reflect what is going on. Ample new documentation is added, as code comments, to explain how things work.

Note that some of the defaults are unexpected, as they are coming from the oniro.git pipeline, and eventually will be included. This can be seen as variables defined on the .oniro-repo-workspace job that are subsequently re-defined in the build job.

Signed-off-by: Zygmunt Krynicki zygmunt.krynicki@huawei.com

.gitlab-ci.yml

@@ -32,45 +32,73 @@ reuse:
 #   Oniro specific variables used during CI/CD process. Variables in this group
 #   should use meaningful names and avoid abbreviations, if possible.
 #
-# CI_ONIRO_BB_LOCAL_CONF_*:
-#   Mechanism for injecting variables into Bitbake configuration files.
-#   Variables in this group represent existing Bitbake/Yocto variables and
-#   retain their original name, apart from the prefix.
+# CI_ONIRO_MANIFEST_REPO_{URL,REV}:
+#   URL and revision of the manifest repository.
+#
+# CI_ONIRO_MANIFEST_MIRROR_REPO_{URL,REV,DIR}:
+#   URL, revision and directory path of the mirror of manifest repository.
+#   This repository is used to speed up construction of repo workspace.
+#
+# CI_ONIRO_REPO_WORKSPACE_DIR:
+#   Directory path of repo workspace.
 #
 # CI_*:
 #   Third party variables used during CI/CD process, defined by GitLab.
 #   Variables in this group are defined by GitLab and retain their original
 #   name.
 #
-# GIT_STRATEGY, CACHE_COMPRESSOIN_LEVEL:
+# GIT_STRATEGY, CACHE_COMPRESSION_LEVEL:
 #   Part of GitLab interface.
 .oniro-repo-workspace:
   interruptible: true
   image:
     name: registry.ostc-eu.org/ostc/oniro/bitbake-builder:latest
   variables:
-    GIT_STRATEGY: none
+    # URL and branch or revision of the oniro.git repository which contains a
+    # repo manifest file.
+    #
+    # The revision should be bumped during the major release of Oniro but both
+    # variables can be changed to CI_PROJECT_URL/CI_COMMIT_SHA when testing
+    # changes landing to oniro.git.
+    CI_ONIRO_MANIFEST_REPO_URL: https://gitlab.eclipse.org/eclipse/oniro-core/oniro.git
+    CI_ONIRO_MANIFEST_REPO_REV: kirkstone
+    # URL and branch used with repo "repo --mirror" to speed up workspace
+    # construction.
+    #
+    # Those are distinct from CI_ONIRO_MANIFEST_REPO_{URL,REV} because the
+    # former variables can be redirected to CI_PROJECT_URL and CI_COMMIT_SHA,
+    # while those two stay fixed.
+    #
+    # The revision should _only_ be bumped during the major release of Oniro.
+    CI_ONIRO_MANIFEST_MIRROR_REPO_URL: https://gitlab.eclipse.org/eclipse/oniro-core/oniro.git
+    CI_ONIRO_MANIFEST_MIRROR_REPO_REV: kirkstone
+    # Directory where repo mirror is constructed. This location is covered by
+    # GitLab cache system, and will be reused between pipelines of the same
+    # project.
+    CI_ONIRO_MANIFEST_MIRROR_REPO_DIR: $CI_PROJECT_DIR/.cache/repo-mirror
+    # Directory where repo workspace is constructed.
+    CI_ONIRO_REPO_WORKSPACE_DIR: $CI_PROJECT_DIR/.tmp/repo-workspace
+    # Use fastest cache compression algorithm, as bulk of the cache is
+    # already-compressed git history.
     CACHE_COMPRESSION_LEVEL: fastest
-    CI_ONIRO_REPO_WORKSPACE_URL: https://gitlab.eclipse.org/eclipse/oniro-core/oniro.git
-    CI_ONIRO_REPO_WORKSPACE_REV: kirkstone
-    CI_ONIRO_REPO_WORKSPACE_DIR: $CI_PROJECT_DIR/.cache/repo-workspace
-    CI_ONIRO_REPO_WORKSPACE_CACHE_PREFIX: workspace
-    CI_ONIRO_REPO_WORKSPACE_LOCAL_MANIFEST: >
-      <?xml version="1.0" encoding="UTF-8"?>
-      <manifest>
-        <!-- remove original docs project entry -->
-        <remove-project name="oniro-core/docs.git" />
-        <!-- add remote representing the project -->
-        <remote name="oniro-override" fetch="${CI_PROJECT_URL}/../" />
-        <!-- add docs at the exact version are testing -->
-        <project name="${CI_PROJECT_NAME}" path="docs" remote="oniro-override" revision="${CI_COMMIT_SHA}" />
-      </manifest>
+    # Ask GitLab _not_ to check out the git repository associated with the
+    # project. This is, in a way, pointless, since we use repo, not pure git,
+    # to construct the workspace. Due to the fact that oniro is
+    # self-referential (the manifest refers to the repository that contains the
+    # manifest). This requires custom logic to behave correctly in scenarios
+    # that modify oniro.git in any way (e.g. a branch, a pull request or merge
+    # train).
+    GIT_STRATEGY: none
   cache:
-    - key: $CI_ONIRO_REPO_WORKSPACE_CACHE_PREFIX-$CI_ONIRO_REPO_WORKSPACE_REV
-      paths: [$CI_ONIRO_REPO_WORKSPACE_DIR]
-      when: always
+    - key:
+        prefix: repo-mirror-$CI_ONIRO_MANIFEST_MIRROR_REPO_REV
+        files:
+          - default.xml
+      paths:
+        - $CI_ONIRO_MANIFEST_MIRROR_REPO_DIR
   before_script:
     - |
+      # Define helper functions to generate GitLab fold markers.
 
       function gl_section_open() {
         printf '\e[0K''section_start'':%s:%s\r\e[0K%s\n' "$(date +%s)" "$1" "$2"
@@ -84,6 +112,9 @@ reuse:
         printf '\e[0K''section_end'':%s:%s\r\e[0K\n' "$(date +%s)" "$1"
       }
 
+      # Query system information. This is mostly useful for forensics, when
+      # something goes wrong and access to basic information of this type can
+      # help to uncover the problem.
     - gl_section_open_collapsed system_info "Querying system information"
     - id
     - uname -a
@@ -93,23 +124,44 @@ reuse:
     - env | grep -E '^CI_ONIRO' | sort
     - gl_section_close system_info
 
+    # Set up Git with bot identity. Eclipse ECA check allows this user to
+    # create and send commits.
     - gl_section_open_collapsed setup_git "Setting up git"
     - git config --global --add safe.directory "$CI_PROJECT_DIR"
     - git config --global user.name "Oniro Core Project Bot"
     - git config --global user.email "oniro-core-bot@eclipse.org"
     - gl_section_close setup_git
 
-    - gl_section_open_collapsed workspace_setup "Setting up repo workspace"
+    # Create and update a mirror for repo, using the semi-fixed manifest mirror
+    # repo URL and revision. Since this is cached, the "repo init" part is
+    # rarely executed (see the test command below), and only the forced
+    # synchronization is executed.
+    #
+    # Note that the location of the mirror is stored in GitLab cache using the
+    # repo revision as cache key, allowing multiple releases to co-exist
+    # efficiently.
+    - gl_section_open_collapsed repo_mirror_setup "Setting up repo mirror"
+    - mkdir -p "$CI_ONIRO_MANIFEST_MIRROR_REPO_DIR"
+    - pushd "$CI_ONIRO_MANIFEST_MIRROR_REPO_DIR"
+    - echo "Initializing repository mirror from $CI_ONIRO_MANIFEST_MIRROR_REPO_URL and $CI_ONIRO_MANIFEST_MIRROR_REPO_REV"
+    - test ! -e .repo && repo init --mirror --manifest-url "$CI_ONIRO_MANIFEST_MIRROR_REPO_URL" --manifest-branch "$CI_ONIRO_MANIFEST_MIRROR_REPO_REV" --no-clone-bundle
+    - echo "Synchronizing repository mirror"
+    - repo sync --force-sync || repo status
+    - gl_section_close repo_mirror_setup
+
+      # Create a repo workspace using the mirror as reference. This is fairly
+      # efficient, as repo will hardlink files (assuming they live on the same
+      # filesystem) and avoid bulk of the network traffic.
+    - gl_section_open_collapsed repo_workspace_setup "Setting up repo workspace"
     - mkdir -p "$CI_ONIRO_REPO_WORKSPACE_DIR"
     - pushd "$CI_ONIRO_REPO_WORKSPACE_DIR"
-    - echo "Initializing repository workspace from $CI_ONIRO_REPO_WORKSPACE_URL and $CI_ONIRO_REPO_WORKSPACE_REV"
-    - repo init --manifest-url "$CI_ONIRO_REPO_WORKSPACE_URL" --manifest-branch "$CI_ONIRO_REPO_WORKSPACE_REV" --no-clone-bundle
+    - echo "Initializing repository workspace from $CI_ONIRO_MANIFEST_REPO_URL and $CI_ONIRO_MANIFEST_REPO_REV"
+    - repo init --reference "$CI_ONIRO_MANIFEST_MIRROR_REPO_DIR" --manifest-url "$CI_ONIRO_MANIFEST_REPO_URL" --manifest-branch "$CI_ONIRO_MANIFEST_REPO_REV" --no-clone-bundle
     - mkdir -p "${CI_ONIRO_REPO_WORKSPACE_DIR}/.repo/local_manifests"
-    - test -n "${CI_ONIRO_REPO_WORKSPACE_LOCAL_MANIFEST-}" && echo "$CI_ONIRO_REPO_WORKSPACE_LOCAL_MANIFEST" | tee "${CI_ONIRO_REPO_WORKSPACE_DIR}/.repo/local_manifests/local.xml"
+    - test -n "${CI_ONIRO_REPO_WORKSPACE_LOCAL_MANIFEST:-}" && echo "$CI_ONIRO_REPO_WORKSPACE_LOCAL_MANIFEST" | tee "${CI_ONIRO_REPO_WORKSPACE_DIR}/.repo/local_manifests/local.xml"
     - echo "Synchronizing repository workspace"
-    - repo sync --force-sync
-    - gl_section_close workspace_setup
-
+    - repo sync --force-sync || repo status
+    - gl_section_close repo_workspace_setup
 
 build-docs:
   extends: [.oniro-repo-workspace]
@@ -122,8 +174,26 @@ build-docs:
   artifacts:
     paths:
       - build
+  variables:
+    # When the workspace is created, substitute the "docs" repository that
+    # described by the manifest with the project being tested. This works for
+    # forks and branches but not for merge requests. For that look at the build
+    # rule below.
+    CI_ONIRO_REPO_WORKSPACE_LOCAL_MANIFEST: >
+      <?xml version="1.0" encoding="UTF-8"?>
+      <manifest>
+        <!-- remove original docs project entry -->
+        <remove-project name="oniro-core/docs.git" />
+        <!-- add remote representing the project -->
+        <remote name="oniro-override" fetch="${CI_PROJECT_URL}/../" />
+        <!-- add docs at the exact version are testing -->
+        <project name="${CI_PROJECT_NAME}" path="docs" remote="oniro-override" revision="${CI_COMMIT_SHA}" />
+      </manifest>
   rules:
-    # Build the docs when a merge request is created.
+    # During the merge request, substitute the "docs" repository that is
+    # described by the manifest with the project that is the source of the
+    # merge request. This does not test the merged result but is the next best
+    # thing we can do right now.
     - if: '$CI_PIPELINE_SOURCE == "merge_request_event"'
       variables:
         CI_ONIRO_REPO_WORKSPACE_LOCAL_MANIFEST: >